[Pig-dev] dataflow in logical plan

On Jan 24, 2011, at 12:48 PM, Baraa Mohamad wrote:

Hello all:

I'm new user of Pig , and I'm very interested in the architecture of
Pig.
I have a question about the logical plan

In the logical plan of this example: (in attach)
a = load 'myfile';
b = filter a by $0 > 5;
store b into 'myfilteredfile';


Does all the data in 'myfile' will be sent in it's totality to the
Proj(0) operator and to the Filter Operator ??
More generally what are runing on the arrows in the logical plan ??

what is the best documentation to understand the architecture of Pig
not only how to use it because I'll try to use it in the medical
domain but first I have to understand it
deeply

thank you very much for your help


Baraa MOHAMAD
Doctorante en informatique
ISIMA-LIMOS
Université Blaise Pascal
Clermont-Ferrand
France
Tél: +33 658900080

On Mon, Jan 24, 2011 at 10:08 PM, Alan Gates wrote:

The logical plan for your script will look like:

Load -> Filter -> Store

Filter will have an expression plan that looks like Proj($0) > const(5)

So yes, all your data will go through the filter operator. But keep in
mind that there is a filter operator in each map task, so all your code will
not go through any one instance of the operator (unless myfile is small).
Hope that helps.

Unfortunately, there is not any great architecture document on Pig.
Probably the best substitute is a paper we published in VLDB 2009, which
you can get here:
http://infolab.stanford.edu/~olston/publications/vldb09.pdf. Since this
is almost 2 years old now some of the specific information is out of date
but the basic structure is still correct.

Alan.


On Jan 24, 2011, at 12:48 PM, Baraa Mohamad wrote:

Hello all:

On Jan 24, 2011, at 1:23 PM, Baraa Mohamad wrote:

Thank you very much for your explination ,
Just to verify that I understood correctly
For example if myfile contains the following data
1 3 4
3 4 6
7 8 2
4 5 9
9 3 5
6 6 2

so all this data will be sent to Proj(0) operator which gives as a
results
1
3
7
4
9
6

After that all this data in myfile will be sent to the filter
operator, so that the filter take tow inputs the myfile data and the
result of the proj(0) > 5 which is
7
9
6

regards


On Mon, Jan 24, 2011 at 10:08 PM, Alan Gates wrote:
The logical plan for your script will look like:

Load -> Filter -> Store

Filter will have an expression plan that looks like Proj($0) >
const(5)

So yes, all your data will go through the filter operator. But keep
in mind that there is a filter operator in each map task, so all
your code will not go through any one instance of the operator
(unless myfile is small). Hope that helps.

Unfortunately, there is not any great architecture document on Pig.
Probably the best substitute is a paper we published in VLDB 2009,
which you can get here: http://infolab.stanford.edu/~olston/publications/vldb09.pdf
. Since this is almost 2 years old now some of the specific
information is out of date but the basic structure is still correct.

Alan.


On Jan 24, 2011, at 12:48 PM, Baraa Mohamad wrote:

Hello all:

I'm new user of Pig , and I'm very interested in the architecture of
Pig.
I have a question about the logical plan

In the logical plan of this example: (in attach)
a = load 'myfile';
b = filter a by $0 > 5;
store b into 'myfilteredfile';


Does all the data in 'myfile' will be sent in it's totality to the
Proj(0) operator and to the Filter Operator ??
More generally what are runing on the arrows in the logical plan ??

what is the best documentation to understand the architecture of Pig
not only how to use it because I'll try to use it in the medical
domain but first I have to understand it
deeply

thank you very much for your help


Baraa MOHAMAD
Doctorante en informatique
ISIMA-LIMOS
Université Blaise Pascal
Clermont-Ferrand
France
Tél: +33 658900080

On Mon, Jan 24, 2011 at 10:32 PM, Alan Gates wrote:

Pig has two levels of operators in its logical (and physical) plans,
relational and expression. Projection is an expression operator in Pig, not
a relational operator (as it is in most databases). So (ignoring the
affects of the optimizer for now) all of your data will be sent to the
filter relational operator. Your filter will see 1 3 4 etc., not 1 etc.
Inside that filter the tuples will be trimmed by the projection operator as
part of the expression plan for '>'.

Alan.


On Jan 24, 2011, at 1:23 PM, Baraa Mohamad wrote:

Thank you very much for your explination ,